Ask anything. About your business, not the internet.

Grounded, not generative

Maya's answers come from tools that query your actual records — never from a model's imagination. When the data does not exist, she says so plainly instead of inventing a plausible number. In accounting, refusing to guess is a feature.

The reach

Sales KPIs, revenue trends, top and slow items, payment and channel breakdowns, forecasts for any metric up to 90 days, stock levels and movements, purchase orders, recipes and reorder plans, expiring items, data health, savings — and exports of any of it to Excel, CSV or PDF from inside the chat.

She remembers

Tell her you close Mondays or that your target margin is 40%, and every future conversation knows. What she remembers is visible and deletable — memory with a settings page, not a black box.

Why grounded beats fluent

A language model asked about your business will produce a confident, well-written answer whether or not it knows anything. That is the failure mode that makes AI unusable in accounting: the wrong number arrives in the same tone as the right one, and nothing in the sentence signals which you got.

Maya answers by calling tools that query your actual records, then reports what came back. When the data does not exist she says so rather than filling the gap. Refusing to guess is not a limitation here — it is the whole reason the answer is worth reading.

What she cannot do

She does not know anything about the world outside your books. Market conditions, what a competitor charges, why footfall dropped — those are questions for a person. She can tell you that Tuesday was down 18% against last Tuesday and which categories moved; she cannot tell you it rained.

Maya is in early access, which means the integration with SeroBooks is still being built rather than generally available. Everything described here exists in the product she runs on today; what is still in progress is bringing her fully inside these books.